A year ago I didn’t get sense that AI was that helpful when coding. It did a lot of mistakes and felt like fixing the mistakes often took more time than building from scratch without AI. I used Cursor’s tab complete pretty much row at a time rather than building bigger features.
Nowadays I write maybe 90% of code using AI. Claude feels better for building bigger features, especially with Fable 5. OpenCode is really good for smaller things and I love how easy it is to use pretty much any models.
I’m still using cloud LLMs but I believe that 1-2 years from now, I’ll have my home server which I use the same way I use OpenCode at the moment. Some bigger features might make sense to write with the best models but the models Apple Studio or Nvidia DGX Spark can run are already pretty amazing.
It’s so impressive how far we have gotten in a year. I don’t think it’s obvious for many other people than software engineers because a lot of the development has impacted the most to our type of job. I’m still very bullish that this development phase will continue. We have so much more to squeeze out of agentic LLMs and as the models get better, the less human intervention is needed.
I remember thinking 2-3 years ago that it’s possible I’m not going to have a job in 5 years. I still think the same way. At the moment some special knowledge is needed and I believe it to stay this way for at least a year but probably longer. Less people are needed and the people who get to keep their work need to adapt to the new reality. But by 2030, I believe that most software that used to require huge teams to maintain can be developed by only one person.
So there will be some need for humans but the field has so many people that it’s like all of them trying to run through a tiny door. It’s just not big enough.
But this is great opportunity to build new startups. I’m not feeling like my world is going to change to worse but I’m excited to adapt to the changes. Smart people always find a way to create and capture value.